fs/ceph/osdmap.c

   1
   2 #include <asm/div64.h>
   3
   4 #include "super.h"
   5 #include "osdmap.h"
   6 #include "crush/hash.h"
   7 #include "crush/mapper.h"
   8 #include "decode.h"
   9 #include "ceph_debug.h"
  10
  11 char *ceph_osdmap_state_str(char *str, int len, int state)
  12 {
  13         int flag = 0;
  14
  15         if (!len)
  16                 goto done;
  17
  18         *str = '\0';
  19         if (state) {
  20                 if (state & CEPH_OSD_EXISTS) {
  21                         snprintf(str, len, "exists");
  22                         flag = 1;
  23                 }
  24                 if (state & CEPH_OSD_UP) {
  25                         snprintf(str, len, "%s%s%s", str, (flag ? ", " : ""),
  26                                  "up");
  27                         flag = 1;
  28                 }
  29         } else {
  30                 snprintf(str, len, "doesn't exist");
  31         }
  32 done:
  33         return str;
  34 }
  35
  36 /* maps */
  37
  38 static int calc_bits_of(unsigned t)
  39 {
  40         int b = 0;
  41         while (t) {
  42                 t = t >> 1;
  43                 b++;
  44         }
  45         return b;
  46 }
  47
  48 /*
  49  * the foo_mask is the smallest value 2^n-1 that is >= foo.
  50  */
  51 static void calc_pg_masks(struct ceph_pg_pool_info *pi)
  52 {
  53         pi->pg_num_mask = (1 << calc_bits_of(le32_to_cpu(pi->v.pg_num)-1)) - 1;
  54         pi->pgp_num_mask =
  55                 (1 << calc_bits_of(le32_to_cpu(pi->v.pgp_num)-1)) - 1;
  56         pi->lpg_num_mask =
  57                 (1 << calc_bits_of(le32_to_cpu(pi->v.lpg_num)-1)) - 1;
  58         pi->lpgp_num_mask =
  59                 (1 << calc_bits_of(le32_to_cpu(pi->v.lpgp_num)-1)) - 1;
  60 }
  61
  62 /*
  63  * decode crush map
  64  */
  65 static int crush_decode_uniform_bucket(void **p, void *end,
  66                                        struct crush_bucket_uniform *b)
  67 {
  68         dout("crush_decode_uniform_bucket %p to %p\n", *p, end);
  69         ceph_decode_need(p, end, (1+b->h.size) * sizeof(u32), bad);
  70         b->item_weight = ceph_decode_32(p);
  71         return 0;
  72 bad:
  73         return -EINVAL;
  74 }
  75
  76 static int crush_decode_list_bucket(void **p, void *end,
  77                                     struct crush_bucket_list *b)
  78 {
  79         int j;
  80         dout("crush_decode_list_bucket %p to %p\n", *p, end);
  81         b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
  82         if (b->item_weights == NULL)
  83                 return -ENOMEM;
  84         b->sum_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
  85         if (b->sum_weights == NULL)
  86                 return -ENOMEM;
  87         ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
  88         for (j = 0; j < b->h.size; j++) {
  89                 b->item_weights[j] = ceph_decode_32(p);
  90                 b->sum_weights[j] = ceph_decode_32(p);
  91         }
  92         return 0;
  93 bad:
  94         return -EINVAL;
  95 }
  96
  97 static int crush_decode_tree_bucket(void **p, void *end,
  98                                     struct crush_bucket_tree *b)
  99 {
 100         int j;
 101         dout("crush_decode_tree_bucket %p to %p\n", *p, end);
 102         ceph_decode_32_safe(p, end, b->num_nodes, bad);
 103         b->node_weights = kcalloc(b->num_nodes, sizeof(u32), GFP_NOFS);
 104         if (b->node_weights == NULL)
 105                 return -ENOMEM;
 106         ceph_decode_need(p, end, b->num_nodes * sizeof(u32), bad);
 107         for (j = 0; j < b->num_nodes; j++)
 108                 b->node_weights[j] = ceph_decode_32(p);
 109         return 0;
 110 bad:
 111         return -EINVAL;
 112 }
 113
 114 static int crush_decode_straw_bucket(void **p, void *end,
 115                                      struct crush_bucket_straw *b)
 116 {
 117         int j;
 118         dout("crush_decode_straw_bucket %p to %p\n", *p, end);
 119         b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
 120         if (b->item_weights == NULL)
 121                 return -ENOMEM;
 122         b->straws = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
 123         if (b->straws == NULL)
 124                 return -ENOMEM;
 125         ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
 126         for (j = 0; j < b->h.size; j++) {
 127                 b->item_weights[j] = ceph_decode_32(p);
 128                 b->straws[j] = ceph_decode_32(p);
 129         }
 130         return 0;
 131 bad:
 132         return -EINVAL;
 133 }
 134
 135 static struct crush_map *crush_decode(void *pbyval, void *end)
 136 {
 137         struct crush_map *c;
 138         int err = -EINVAL;
 139         int i, j;
 140         void **p = &pbyval;
 141         void *start = pbyval;
 142         u32 magic;
 143
 144         dout("crush_decode %p to %p len %d\n", *p, end, (int)(end - *p));
 145
 146         c = kzalloc(sizeof(*c), GFP_NOFS);
 147         if (c == NULL)
 148                 return ERR_PTR(-ENOMEM);
 149
 150         ceph_decode_need(p, end, 4*sizeof(u32), bad);
 151         magic = ceph_decode_32(p);
 152         if (magic != CRUSH_MAGIC) {
 153                 pr_err("crush_decode magic %x != current %x\n",
 154                        (unsigned)magic, (unsigned)CRUSH_MAGIC);
 155                 goto bad;
 156         }
 157         c->max_buckets = ceph_decode_32(p);
 158         c->max_rules = ceph_decode_32(p);
 159         c->max_devices = ceph_decode_32(p);
 160
 161         c->device_parents = kcalloc(c->max_devices, sizeof(u32), GFP_NOFS);
 162         if (c->device_parents == NULL)
 163                 goto badmem;
 164         c->bucket_parents = kcalloc(c->max_buckets, sizeof(u32), GFP_NOFS);
 165         if (c->bucket_parents == NULL)
 166                 goto badmem;
 167
 168         c->buckets = kcalloc(c->max_buckets, sizeof(*c->buckets), GFP_NOFS);
 169         if (c->buckets == NULL)
 170                 goto badmem;
 171         c->rules = kcalloc(c->max_rules, sizeof(*c->rules), GFP_NOFS);
 172         if (c->rules == NULL)
 173                 goto badmem;
 174
 175         /* buckets */
 176         for (i = 0; i < c->max_buckets; i++) {
 177                 int size = 0;
 178                 u32 alg;
 179                 struct crush_bucket *b;
 180
 181                 ceph_decode_32_safe(p, end, alg, bad);
 182                 if (alg == 0) {
 183                         c->buckets[i] = NULL;
 184                         continue;
 185                 }
 186                 dout("crush_decode bucket %d off %x %p to %p\n",
 187                      i, (int)(*p-start), *p, end);
 188
 189                 switch (alg) {
 190                 case CRUSH_BUCKET_UNIFORM:
 191                         size = sizeof(struct crush_bucket_uniform);
 192                         break;
 193                 case CRUSH_BUCKET_LIST:
 194                         size = sizeof(struct crush_bucket_list);
 195                         break;
 196                 case CRUSH_BUCKET_TREE:
 197                         size = sizeof(struct crush_bucket_tree);
 198                         break;
 199                 case CRUSH_BUCKET_STRAW:
 200                         size = sizeof(struct crush_bucket_straw);
 201                         break;
 202                 default:
 203                         goto bad;
 204                 }
 205                 BUG_ON(size == 0);
 206                 b = c->buckets[i] = kzalloc(size, GFP_NOFS);
 207                 if (b == NULL)
 208                         goto badmem;
 209
 210                 ceph_decode_need(p, end, 4*sizeof(u32), bad);
 211                 b->id = ceph_decode_32(p);
 212                 b->type = ceph_decode_16(p);
 213                 b->alg = ceph_decode_16(p);
 214                 b->weight = ceph_decode_32(p);
 215                 b->size = ceph_decode_32(p);
 216
 217                 dout("crush_decode bucket size %d off %x %p to %p\n",
 218                      b->size, (int)(*p-start), *p, end);
 219
 220                 b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS);
 221                 if (b->items == NULL)
 222                         goto badmem;
 223                 b->perm = kcalloc(b->size, sizeof(u32), GFP_NOFS);
 224                 if (b->perm == NULL)
 225                         goto badmem;
 226                 b->perm_n = 0;
 227
 228                 ceph_decode_need(p, end, b->size*sizeof(u32), bad);
 229                 for (j = 0; j < b->size; j++)
 230                         b->items[j] = ceph_decode_32(p);
 231
 232                 switch (b->alg) {
 233                 case CRUSH_BUCKET_UNIFORM:
 234                         err = crush_decode_uniform_bucket(p, end,
 235                                   (struct crush_bucket_uniform *)b);
 236                         if (err < 0)
 237                                 goto bad;
 238                         break;
 239                 case CRUSH_BUCKET_LIST:
 240                         err = crush_decode_list_bucket(p, end,
 241                                (struct crush_bucket_list *)b);
 242                         if (err < 0)
 243                                 goto bad;
 244                         break;
 245                 case CRUSH_BUCKET_TREE:
 246                         err = crush_decode_tree_bucket(p, end,
 247                                 (struct crush_bucket_tree *)b);
 248                         if (err < 0)
 249                                 goto bad;
 250                         break;
 251                 case CRUSH_BUCKET_STRAW:
 252                         err = crush_decode_straw_bucket(p, end,
 253                                 (struct crush_bucket_straw *)b);
 254                         if (err < 0)
 255                                 goto bad;
 256                         break;
 257                 }
 258         }
 259
 260         /* rules */
 261         dout("rule vec is %p\n", c->rules);
 262         for (i = 0; i < c->max_rules; i++) {
 263                 u32 yes;
 264                 struct crush_rule *r;
 265
 266                 ceph_decode_32_safe(p, end, yes, bad);
 267                 if (!yes) {
 268                         dout("crush_decode NO rule %d off %x %p to %p\n",
 269                              i, (int)(*p-start), *p, end);
 270                         c->rules[i] = NULL;
 271                         continue;
 272                 }
 273
 274                 dout("crush_decode rule %d off %x %p to %p\n",
 275                      i, (int)(*p-start), *p, end);
 276
 277                 /* len */
 278                 ceph_decode_32_safe(p, end, yes, bad);
 279 #if BITS_PER_LONG == 32
 280                 if (yes > ULONG_MAX / sizeof(struct crush_rule_step))
 281                         goto bad;
 282 #endif
 283                 r = c->rules[i] = kmalloc(sizeof(*r) +
 284                                           yes*sizeof(struct crush_rule_step),
 285                                           GFP_NOFS);
 286                 if (r == NULL)
 287                         goto badmem;
 288                 dout(" rule %d is at %p\n", i, r);
 289                 r->len = yes;
 290                 ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
 291                 ceph_decode_need(p, end, r->len*3*sizeof(u32), bad);
 292                 for (j = 0; j < r->len; j++) {
 293                         r->steps[j].op = ceph_decode_32(p);
 294                         r->steps[j].arg1 = ceph_decode_32(p);
 295                         r->steps[j].arg2 = ceph_decode_32(p);
 296                 }
 297         }
 298
 299         /* ignore trailing name maps. */
 300
 301         dout("crush_decode success\n");
 302         return c;
 303
 304 badmem:
 305         err = -ENOMEM;
 306 bad:
 307         dout("crush_decode fail %d\n", err);
 308         crush_destroy(c);
 309         return ERR_PTR(err);
 310 }
 311
 312
 313 /*
 314  * osd map
 315  */
 316 void ceph_osdmap_destroy(struct ceph_osdmap *map)
 317 {
 318         dout("osdmap_destroy %p\n", map);
 319         if (map->crush)
 320                 crush_destroy(map->crush);
 321         while (!RB_EMPTY_ROOT(&map->pg_temp))
 322                 rb_erase(rb_first(&map->pg_temp), &map->pg_temp);
 323         kfree(map->osd_state);
 324         kfree(map->osd_weight);
 325         kfree(map->pg_pool);
 326         kfree(map->osd_addr);
 327         kfree(map);
 328 }
 329
 330 /*
 331  * adjust max osd value.  reallocate arrays.
 332  */
 333 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
 334 {
 335         u8 *state;
 336         struct ceph_entity_addr *addr;
 337         u32 *weight;
 338
 339         state = kcalloc(max, sizeof(*state), GFP_NOFS);
 340         addr = kcalloc(max, sizeof(*addr), GFP_NOFS);
 341         weight = kcalloc(max, sizeof(*weight), GFP_NOFS);
 342         if (state == NULL || addr == NULL || weight == NULL) {
 343                 kfree(state);
 344                 kfree(addr);
 345                 kfree(weight);
 346                 return -ENOMEM;
 347         }
 348
 349         /* copy old? */
 350         if (map->osd_state) {
 351                 memcpy(state, map->osd_state, map->max_osd*sizeof(*state));
 352                 memcpy(addr, map->osd_addr, map->max_osd*sizeof(*addr));
 353                 memcpy(weight, map->osd_weight, map->max_osd*sizeof(*weight));
 354                 kfree(map->osd_state);
 355                 kfree(map->osd_addr);
 356                 kfree(map->osd_weight);
 357         }
 358
 359         map->osd_state = state;
 360         map->osd_weight = weight;
 361         map->osd_addr = addr;
 362         map->max_osd = max;
 363         return 0;
 364 }
 365
 366 /*
 367  * Insert a new pg_temp mapping
 368  */
 369 static int pgid_cmp(struct ceph_pg l, struct ceph_pg r)
 370 {
 371         u64 a = *(u64 *)&l;
 372         u64 b = *(u64 *)&r;
 373
 374         if (a < b)
 375                 return -1;
 376         if (a > b)
 377                 return 1;
 378         return 0;
 379 }
 380
 381 static int __insert_pg_mapping(struct ceph_pg_mapping *new,
 382                                struct rb_root *root)
 383 {
 384         struct rb_node **p = &root->rb_node;
 385         struct rb_node *parent = NULL;
 386         struct ceph_pg_mapping *pg = NULL;
 387         int c;
 388
 389         while (*p) {
 390                 parent = *p;
 391                 pg = rb_entry(parent, struct ceph_pg_mapping, node);
 392                 c = pgid_cmp(new->pgid, pg->pgid);
 393                 if (c < 0)
 394                         p = &(*p)->rb_left;
 395                 else if (c > 0)
 396                         p = &(*p)->rb_right;
 397                 else
 398                         return -EEXIST;
 399         }
 400
 401         rb_link_node(&new->node, parent, p);
 402         rb_insert_color(&new->node, root);
 403         return 0;
 404 }
 405
 406 /*
 407  * decode a full map.
 408  */
 409 struct ceph_osdmap *osdmap_decode(void **p, void *end)
 410 {
 411         struct ceph_osdmap *map;
 412         u16 version;
 413         u32 len, max, i;
 414         int err = -EINVAL;
 415         void *start = *p;
 416
 417         dout("osdmap_decode %p to %p len %d\n", *p, end, (int)(end - *p));
 418
 419         map = kzalloc(sizeof(*map), GFP_NOFS);
 420         if (map == NULL)
 421                 return ERR_PTR(-ENOMEM);
 422         map->pg_temp = RB_ROOT;
 423
 424         ceph_decode_16_safe(p, end, version, bad);
 425
 426         ceph_decode_need(p, end, 2*sizeof(u64)+6*sizeof(u32), bad);
 427         ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
 428         map->epoch = ceph_decode_32(p);
 429         ceph_decode_copy(p, &map->created, sizeof(map->created));
 430         ceph_decode_copy(p, &map->modified, sizeof(map->modified));
 431
 432         map->num_pools = ceph_decode_32(p);
 433         map->pg_pool = kcalloc(map->num_pools, sizeof(*map->pg_pool),
 434                                GFP_NOFS);
 435         if (!map->pg_pool) {
 436                 err = -ENOMEM;
 437                 goto bad;
 438         }
 439         ceph_decode_32_safe(p, end, max, bad);
 440         while (max--) {
 441                 ceph_decode_need(p, end, 4+sizeof(map->pg_pool->v), bad);
 442                 i = ceph_decode_32(p);
 443                 if (i >= map->num_pools)
 444                         goto bad;
 445                 ceph_decode_copy(p, &map->pg_pool[i].v,
 446                                  sizeof(map->pg_pool->v));
 447                 calc_pg_masks(&map->pg_pool[i]);
 448                 p += le32_to_cpu(map->pg_pool[i].v.num_snaps) * sizeof(u64);
 449                 p += le32_to_cpu(map->pg_pool[i].v.num_removed_snap_intervals)
 450                         * sizeof(u64) * 2;
 451         }
 452
 453         ceph_decode_32_safe(p, end, map->flags, bad);
 454
 455         max = ceph_decode_32(p);
 456
 457         /* (re)alloc osd arrays */
 458         err = osdmap_set_max_osd(map, max);
 459         if (err < 0)
 460                 goto bad;
 461         dout("osdmap_decode max_osd = %d\n", map->max_osd);
 462
 463         /* osds */
 464         err = -EINVAL;
 465         ceph_decode_need(p, end, 3*sizeof(u32) +
 466                          map->max_osd*(1 + sizeof(*map->osd_weight) +
 467                                        sizeof(*map->osd_addr)), bad);
 468         *p += 4; /* skip length field (should match max) */
 469         ceph_decode_copy(p, map->osd_state, map->max_osd);
 470
 471         *p += 4; /* skip length field (should match max) */
 472         for (i = 0; i < map->max_osd; i++)
 473                 map->osd_weight[i] = ceph_decode_32(p);
 474
 475         *p += 4; /* skip length field (should match max) */
 476         ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
 477         for (i = 0; i < map->max_osd; i++)
 478                 ceph_decode_addr(&map->osd_addr[i]);
 479
 480         /* pg_temp */
 481         ceph_decode_32_safe(p, end, len, bad);
 482         for (i = 0; i < len; i++) {
 483                 int n, j;
 484                 struct ceph_pg pgid;
 485                 struct ceph_pg_mapping *pg;
 486
 487                 ceph_decode_need(p, end, sizeof(u32) + sizeof(u64), bad);
 488                 ceph_decode_copy(p, &pgid, sizeof(pgid));
 489                 n = ceph_decode_32(p);
 490                 ceph_decode_need(p, end, n * sizeof(u32), bad);
 491                 pg = kmalloc(sizeof(*pg) + n*sizeof(u32), GFP_NOFS);
 492                 if (!pg) {
 493                         err = -ENOMEM;
 494                         goto bad;
 495                 }
 496                 pg->pgid = pgid;
 497                 pg->len = n;
 498                 for (j = 0; j < n; j++)
 499                         pg->osds[j] = ceph_decode_32(p);
 500
 501                 err = __insert_pg_mapping(pg, &map->pg_temp);
 502                 if (err)
 503                         goto bad;
 504                 dout(" added pg_temp %llx len %d\n", *(u64 *)&pgid, len);
 505         }
 506
 507         /* crush */
 508         ceph_decode_32_safe(p, end, len, bad);
 509         dout("osdmap_decode crush len %d from off 0x%x\n", len,
 510              (int)(*p - start));
 511         ceph_decode_need(p, end, len, bad);
 512         map->crush = crush_decode(*p, end);
 513         *p += len;
 514         if (IS_ERR(map->crush)) {
 515                 err = PTR_ERR(map->crush);
 516                 map->crush = NULL;
 517                 goto bad;
 518         }
 519
 520         /* ignore the rest of the map */
 521         *p = end;
 522
 523         dout("osdmap_decode done %p %p\n", *p, end);
 524         return map;
 525
 526 bad:
 527         dout("osdmap_decode fail\n");
 528         ceph_osdmap_destroy(map);
 529         return ERR_PTR(err);
 530 }
 531
 532 /*
 533  * decode and apply an incremental map update.
 534  */
 535 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
 536                                              struct ceph_osdmap *map,
 537                                              struct ceph_messenger *msgr)
 538 {
 539         struct ceph_osdmap *newmap = map;
 540         struct crush_map *newcrush = NULL;
 541         struct ceph_fsid fsid;
 542         u32 epoch = 0;
 543         struct ceph_timespec modified;
 544         u32 len, pool;
 545         __s32 new_flags, max;
 546         void *start = *p;
 547         int err = -EINVAL;
 548         u16 version;
 549         struct rb_node *rbp;
 550
 551         ceph_decode_16_safe(p, end, version, bad);
 552
 553         ceph_decode_need(p, end, sizeof(fsid)+sizeof(modified)+2*sizeof(u32),
 554                          bad);
 555         ceph_decode_copy(p, &fsid, sizeof(fsid));
 556         epoch = ceph_decode_32(p);
 557         BUG_ON(epoch != map->epoch+1);
 558         ceph_decode_copy(p, &modified, sizeof(modified));
 559         new_flags = ceph_decode_32(p);
 560
 561         /* full map? */
 562         ceph_decode_32_safe(p, end, len, bad);
 563         if (len > 0) {
 564                 dout("apply_incremental full map len %d, %p to %p\n",
 565                      len, *p, end);
 566                 newmap = osdmap_decode(p, min(*p+len, end));
 567                 return newmap;  /* error or not */
 568         }
 569
 570         /* new crush? */
 571         ceph_decode_32_safe(p, end, len, bad);
 572         if (len > 0) {
 573                 dout("apply_incremental new crush map len %d, %p to %p\n",
 574                      len, *p, end);
 575                 newcrush = crush_decode(*p, min(*p+len, end));
 576                 if (IS_ERR(newcrush))
 577                         return ERR_PTR(PTR_ERR(newcrush));
 578         }
 579
 580         /* new flags? */
 581         if (new_flags >= 0)
 582                 map->flags = new_flags;
 583
 584         ceph_decode_need(p, end, 5*sizeof(u32), bad);
 585
 586         /* new max? */
 587         max = ceph_decode_32(p);
 588         if (max >= 0) {
 589                 err = osdmap_set_max_osd(map, max);
 590                 if (err < 0)
 591                         goto bad;
 592         }
 593
 594         map->epoch++;
 595         map->modified = map->modified;
 596         if (newcrush) {
 597                 if (map->crush)
 598                         crush_destroy(map->crush);
 599                 map->crush = newcrush;
 600                 newcrush = NULL;
 601         }
 602
 603         /* new_pool */
 604         ceph_decode_32_safe(p, end, len, bad);
 605         while (len--) {
 606                 ceph_decode_32_safe(p, end, pool, bad);
 607                 if (pool >= map->num_pools) {
 608                         void *pg_pool = kcalloc(pool + 1,
 609                                                 sizeof(*map->pg_pool),
 610                                                 GFP_NOFS);
 611                         if (!pg_pool) {
 612                                 err = -ENOMEM;
 613                                 goto bad;
 614                         }
 615                         memcpy(pg_pool, map->pg_pool,
 616                                map->num_pools * sizeof(*map->pg_pool));
 617                         kfree(map->pg_pool);
 618                         map->pg_pool = pg_pool;
 619                         map->num_pools = pool+1;
 620                 }
 621                 ceph_decode_copy(p, &map->pg_pool[pool].v,
 622                                  sizeof(map->pg_pool->v));
 623                 calc_pg_masks(&map->pg_pool[pool]);
 624         }
 625
 626         /* old_pool (ignore) */
 627         ceph_decode_32_safe(p, end, len, bad);
 628         *p += len * sizeof(u32);
 629
 630         /* new_up */
 631         err = -EINVAL;
 632         ceph_decode_32_safe(p, end, len, bad);
 633         while (len--) {
 634                 u32 osd;
 635                 struct ceph_entity_addr addr;
 636                 ceph_decode_32_safe(p, end, osd, bad);
 637                 ceph_decode_copy_safe(p, end, &addr, sizeof(addr), bad);
 638                 ceph_decode_addr(&addr);
 639                 pr_info("osd%d up\n", osd);
 640                 BUG_ON(osd >= map->max_osd);
 641                 map->osd_state[osd] |= CEPH_OSD_UP;
 642                 map->osd_addr[osd] = addr;
 643         }
 644
 645         /* new_down */
 646         ceph_decode_32_safe(p, end, len, bad);
 647         while (len--) {
 648                 u32 osd;
 649                 ceph_decode_32_safe(p, end, osd, bad);
 650                 (*p)++;  /* clean flag */
 651                 pr_info("osd%d down\n", osd);
 652                 if (osd < map->max_osd)
 653                         map->osd_state[osd] &= ~CEPH_OSD_UP;
 654         }
 655
 656         /* new_weight */
 657         ceph_decode_32_safe(p, end, len, bad);
 658         while (len--) {
 659                 u32 osd, off;
 660                 ceph_decode_need(p, end, sizeof(u32)*2, bad);
 661                 osd = ceph_decode_32(p);
 662                 off = ceph_decode_32(p);
 663                 pr_info("osd%d weight 0x%x %s\n", osd, off,
 664                      off == CEPH_OSD_IN ? "(in)" :
 665                      (off == CEPH_OSD_OUT ? "(out)" : ""));
 666                 if (osd < map->max_osd)
 667                         map->osd_weight[osd] = off;
 668         }
 669
 670         /* new_pg_temp */
 671         rbp = rb_first(&map->pg_temp);
 672         ceph_decode_32_safe(p, end, len, bad);
 673         while (len--) {
 674                 struct ceph_pg_mapping *pg;
 675                 int j;
 676                 struct ceph_pg pgid;
 677                 u32 pglen;
 678                 ceph_decode_need(p, end, sizeof(u64) + sizeof(u32), bad);
 679                 ceph_decode_copy(p, &pgid, sizeof(pgid));
 680                 pglen = ceph_decode_32(p);
 681
 682                 /* remove any? */
 683                 while (rbp && pgid_cmp(rb_entry(rbp, struct ceph_pg_mapping,
 684                                                 node)->pgid, pgid) <= 0) {
 685                         struct rb_node *cur = rbp;
 686                         rbp = rb_next(rbp);
 687                         dout(" removed pg_temp %llx\n",
 688                              *(u64 *)&rb_entry(cur, struct ceph_pg_mapping,
 689                                                node)->pgid);
 690                         rb_erase(cur, &map->pg_temp);
 691                 }
 692
 693                 if (pglen) {
 694                         /* insert */
 695                         ceph_decode_need(p, end, pglen*sizeof(u32), bad);
 696                         pg = kmalloc(sizeof(*pg) + sizeof(u32)*pglen, GFP_NOFS);
 697                         if (!pg) {
 698                                 err = -ENOMEM;
 699                                 goto bad;
 700                         }
 701                         pg->pgid = pgid;
 702                         pg->len = pglen;
 703                         for (j = 0; j < len; j++)
 704                                 pg->osds[j] = ceph_decode_32(p);
 705                         err = __insert_pg_mapping(pg, &map->pg_temp);
 706                         if (err)
 707                                 goto bad;
 708                         dout(" added pg_temp %llx len %d\n", *(u64 *)&pgid,
 709                              pglen);
 710                 }
 711         }
 712         while (rbp) {
 713                 struct rb_node *cur = rbp;
 714                 rbp = rb_next(rbp);
 715                 dout(" removed pg_temp %llx\n",
 716                      *(u64 *)&rb_entry(cur, struct ceph_pg_mapping,
 717                                        node)->pgid);
 718                 rb_erase(cur, &map->pg_temp);
 719         }
 720
 721         /* ignore the rest */
 722         *p = end;
 723         return map;
 724
 725 bad:
 726         pr_err("corrupt inc osdmap epoch %d off %d (%p of %p-%p)\n",
 727                epoch, (int)(*p - start), *p, start, end);
 728         if (newcrush)
 729                 crush_destroy(newcrush);
 730         return ERR_PTR(err);
 731 }
 732
 733
 734
 735
 736 /*
 737  * calculate file layout from given offset, length.
 738  * fill in correct oid, logical length, and object extent
 739  * offset, length.
 740  *
 741  * for now, we write only a single su, until we can
 742  * pass a stride back to the caller.
 743  */
 744 void ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
 745                                    u64 off, u64 *plen,
 746                                    u64 *ono,
 747                                    u64 *oxoff, u64 *oxlen)
 748 {
 749         u32 osize = le32_to_cpu(layout->fl_object_size);
 750         u32 su = le32_to_cpu(layout->fl_stripe_unit);
 751         u32 sc = le32_to_cpu(layout->fl_stripe_count);
 752         u32 bl, stripeno, stripepos, objsetno;
 753         u32 su_per_object;
 754         u64 t, su_offset;
 755
 756         dout("mapping %llu~%llu  osize %u fl_su %u\n", off, *plen,
 757              osize, su);
 758         su_per_object = osize / su;
 759         dout("osize %u / su %u = su_per_object %u\n", osize, su,
 760              su_per_object);
 761
 762         BUG_ON((su & ~PAGE_MASK) != 0);
 763         /* bl = *off / su; */
 764         t = off;
 765         do_div(t, su);
 766         bl = t;
 767         dout("off %llu / su %u = bl %u\n", off, su, bl);
 768
 769         stripeno = bl / sc;
 770         stripepos = bl % sc;
 771         objsetno = stripeno / su_per_object;
 772
 773         *ono = objsetno * sc + stripepos;
 774         dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned)*ono);
 775
 776         /* *oxoff = *off % layout->fl_stripe_unit;  # offset in su */
 777         t = off;
 778         su_offset = do_div(t, su);
 779         *oxoff = su_offset + (stripeno % su_per_object) * su;
 780
 781         /*
 782          * Calculate the length of the extent being written to the selected
 783          * object. This is the minimum of the full length requested (plen) or
 784          * the remainder of the current stripe being written to.
 785          */
 786         *oxlen = min_t(u64, *plen, su - su_offset);
 787         *plen = *oxlen;
 788
 789         dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
 790 }
 791
 792 /*
 793  * calculate an object layout (i.e. pgid) from an oid,
 794  * file_layout, and osdmap
 795  */
 796 int ceph_calc_object_layout(struct ceph_object_layout *ol,
 797                             const char *oid,
 798                             struct ceph_file_layout *fl,
 799                             struct ceph_osdmap *osdmap)
 800 {
 801         unsigned num, num_mask;
 802         struct ceph_pg pgid;
 803         s32 preferred = (s32)le32_to_cpu(fl->fl_pg_preferred);
 804         int poolid = le32_to_cpu(fl->fl_pg_pool);
 805         struct ceph_pg_pool_info *pool;
 806         unsigned ps;
 807
 808         if (poolid >= osdmap->num_pools)
 809                 return -EIO;
 810
 811         pool = &osdmap->pg_pool[poolid];
 812         ps = ceph_full_name_hash(oid, strlen(oid));
 813         if (preferred >= 0) {
 814                 ps += preferred;
 815                 num = le32_to_cpu(pool->v.lpg_num);
 816                 num_mask = pool->lpg_num_mask;
 817         } else {
 818                 num = le32_to_cpu(pool->v.pg_num);
 819                 num_mask = pool->pg_num_mask;
 820         }
 821
 822         pgid.ps = cpu_to_le16(ps);
 823         pgid.preferred = cpu_to_le16(preferred);
 824         pgid.pool = fl->fl_pg_pool;
 825         if (preferred >= 0)
 826                 dout("calc_object_layout '%s' pgid %d.%xp%d\n", oid, poolid, ps,
 827                      (int)preferred);
 828         else
 829                 dout("calc_object_layout '%s' pgid %d.%x\n", oid, poolid, ps);
 830
 831         ol->ol_pgid = pgid;
 832         ol->ol_stripe_unit = fl->fl_object_stripe_unit;
 833         return 0;
 834 }
 835
 836 /*
 837  * Calculate raw osd vector for the given pgid.  Return pointer to osd
 838  * array, or NULL on failure.
 839  */
 840 static int *calc_pg_raw(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
 841                         int *osds, int *num)
 842 {
 843         struct rb_node *n = osdmap->pg_temp.rb_node;
 844         struct ceph_pg_mapping *pg;
 845         struct ceph_pg_pool_info *pool;
 846         int ruleno;
 847         unsigned poolid, ps, pps;
 848         int preferred;
 849         int c;
 850
 851         /* pg_temp? */
 852         while (n) {
 853                 pg = rb_entry(n, struct ceph_pg_mapping, node);
 854                 c = pgid_cmp(pgid, pg->pgid);
 855                 if (c < 0)
 856                         n = n->rb_left;
 857                 else if (c > 0)
 858                         n = n->rb_right;
 859                 else {
 860                         *num = pg->len;
 861                         return pg->osds;
 862                 }
 863         }
 864
 865         /* crush */
 866         poolid = le32_to_cpu(pgid.pool);
 867         ps = le16_to_cpu(pgid.ps);
 868         preferred = (s16)le16_to_cpu(pgid.preferred);
 869
 870         if (poolid >= osdmap->num_pools)
 871                 return NULL;
 872         pool = &osdmap->pg_pool[poolid];
 873         ruleno = crush_find_rule(osdmap->crush, pool->v.crush_ruleset,
 874                                  pool->v.type, pool->v.size);
 875         if (ruleno < 0) {
 876                 pr_err("no crush rule pool %d type %d size %d\n",
 877                        poolid, pool->v.type, pool->v.size);
 878                 return NULL;
 879         }
 880
 881         if (preferred >= 0)
 882                 pps = ceph_stable_mod(ps,
 883                                       le32_to_cpu(pool->v.lpgp_num),
 884                                       pool->lpgp_num_mask);
 885         else
 886                 pps = ceph_stable_mod(ps,
 887                                       le32_to_cpu(pool->v.pgp_num),
 888                                       pool->pgp_num_mask);
 889         pps += poolid;
 890         *num = crush_do_rule(osdmap->crush, ruleno, pps, osds,
 891                              min_t(int, pool->v.size, *num),
 892                              preferred, osdmap->osd_weight);
 893         return osds;
 894 }
 895
 896 /*
 897  * Return primary osd for given pgid, or -1 if none.
 898  */
 899 int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, struct ceph_pg pgid)
 900 {
 901         int rawosds[10], *osds;
 902         int i, num = ARRAY_SIZE(rawosds);
 903
 904         osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
 905         if (!osds)
 906                 return -1;
 907
 908         /* primary is first up osd */
 909         for (i = 0; i < num; i++)
 910                 if (ceph_osd_is_up(osdmap, osds[i])) {
 911                         return osds[i];
 912                         break;
 913                 }
 914         return -1;
 915 }